MiMo-V2-Pro is a flagship foundation model developed by Xiaomi, released in March 2026. Built specifically for the "agentic era," the model is designed to function as the central reasoning engine for autonomous systems capable of orchestrating complex workflows, software engineering tasks, and long-running production processes. It distinguishes itself from general-purpose chatbots by focusing on multi-step planning, tool use, and environmental interaction, a design philosophy Xiaomi describes as moving from "coding to claw."
The model utilizes a sparse Mixture-of-Experts (MoE) architecture with over 1 trillion total parameters, of which 42 billion are active during any single forward pass. This design allows it to achieve performance levels comparable to the largest proprietary frontier models while maintaining high inference efficiency. A key architectural innovation is its evolved Hybrid Attention mechanism, which uses a 7:1 ratio of Local Sliding Window Attention to Global Attention layers. This configuration enables the model to manage its massive 1-million-token context window with reduced KV cache requirements and lower computational overhead compared to standard quadratic transformers.
In terms of performance, MiMo-V2-Pro excels in agentic benchmarks such as PinchBench and GDPval-AA, which measure a model's ability to execute real-world tasks and browser-based actions. During its development phase, the model was tested anonymously on third-party platforms under the codename "Hunter Alpha," where it briefly topped global usage charts and sparked significant industry speculation before its official identity was revealed.
MiMo-V2-Pro is optimized for long-context reasoning and high-fidelity tool calling. Xiaomi positions the model as the core infrastructure for its broader ecosystem, including integration with mobile OS agents and enterprise document workflows. While its smaller counterpart, MiMo-V2-Flash, was released with open weights, the Pro version is offered primarily as a proprietary model through official API services with tiered pricing based on context usage.